rockenzo

Rock (Ge) Wang · Web Development · Scalable Applications · AI/ML

🧭 About me

const rock = {
  studying:  "Master of IT (Applied AI) @ Adelaide University",
  graduated: "Bachelor of IT @ University of Adelaide (2025)",
  focus:     ["Web Development", "Scalable Applications", "AI / ML"],
  building:  "data pipelines, explainable NLP and local-first AI tools",
  certified: "HackerRank Software Engineer (2024)",
  askMeAbout: ["React & Vue", "PyTorch", "Elasticsearch", "Docker"],
};

⚡ Tech stack

Languages

Frontend & backend

Data & infrastructure

Machine learning & LLMs

🚀 Featured projects

Knia Maps. Geospatial ETL for a Sydney accessibility map using Elasticsearch, PostgreSQL, Redis, Celery and Flask, with reproducible service integration tests.	CSIRO's Data61. Nine-class scam detection with word and character features. Macro F1 0.948 on an 18,261-record internal test grouped by template. Evaluation.
CSIRO's Data61. 194,913 scam and legitimate samples from 19 retained source datasets across email, SMS, popups and dialogue, with verified hashes and reproducible provenance-preserving builds.	Extract structured JSON from webpages with local LLMs, traceable source fields and duplicate merging. Field exact-match F1 0.938 with Qwen on 16 synthetic regression fixtures. Evaluation.
SwiftUI weather app with live location, smooth animations and OpenWeatherMap data.	YOLOv8s football detection, player tracking and automatic pitch mapping, with a static match viewer: momentum timeline, critical moments, tactic notes, per-player distance and speed, and space control replayed over the video and a calibrated pitch. Detector fine-tuned to 0.895 mAP50 (ball recall 0.61) on a small validation set. Experimental estimates from one 30 s clip, not industrial-grade; the ball is still missed about four times in ten. Live demo · Results & limitations.

More projects

- [Odyssey](https://github.com/RockENZO/odyssey): a terminal coding agent with 17 tools, local Ollama inference and optional external web research - [MagicTrick](https://github.com/RockENZO/MagicTrick): an iOS card trick with no buttons, only rotate, shake and a secret tap - [ControllerMouse](https://github.com/RockENZO/ControllerMouse): drive your Mac's cursor with a game controller from the menu bar - [Road semantic segmentation](https://github.com/RockENZO/Road_Semantic_Segmentation_CV): PyTorch segmentation for driving scenes, with mIoU and FLOPs ablations - [Flappy Bird DQN](https://github.com/RockENZO/flappy_bird_dqn-main): a DQN agent that learns to play Flappy Bird - [Feed the World](https://github.com/RockENZO/Feed-the-world-web-app): volunteer-group web app (team project)

🎮 Live playground

Browser demos you can open live. Click a recording to play it.

Roadster and Beyond A fan-made recreation of the Tesla Roadster launch night, with rigid-body physics, a broadcast-style HUD, the four highlight stunts and a day or night track day at Spa-Francorchamps. Keyboard and gamepad.	Touch Grass An interactive 3D meadow: run your hand through 260k instanced blades of grass, then switch between Morning, Golden hour and Night.
Football match analysis YOLOv8s detection and tracking with automatic pitch mapping, replayed as a static viewer: momentum timeline, critical moments, tactic notes, per-player speed and space control. Experimental, from one 30 s clip.

🎓 Education & certification

Master of Information Technology (Applied Artificial Intelligence), Adelaide University (2026 to present).
Bachelor of Information Technology, University of Adelaide (2023 to 2025). High Distinctions in Applied Programming and in Computer Systems, Networks & Security; Distinctions in AI Technologies, Web & Database Computing and Programming for IT Specialists.
HackerRank Software Engineer (Sep 2024): problem solving, Python, SQL, REST APIs.

📊 Development activity

GitHub streak

Contribution snake